Investments, tenth edition

  Anomalies or Data Mining?   

We have covered many of the so-called anomalies 

cited in the literature, but our list could go on and on. Some wonder whether these anoma-

lies are really unexplained puzzles in financial markets, or whether they instead are an 

artifact of data mining. After all, if one reruns the computer database of past returns over 

and over and examines stock returns along enough dimensions, simple chance will cause 

some criteria to  appear  to predict returns. 

 In this regard, it is noteworthy that some anomalies have not shown much staying 

power after being reported in the academic literature. For example, after the small-firm 

effect was published in the early 1980s, it promptly disappeared for much of the rest of 

the decade. 

 Still, even acknowledging the potential for data mining, a common thread seems to 

run through many of the anomalies we have considered, lending support to the notion that 

there is a real puzzle to explain. Value stocks—defined by low P/E ratio, high book-to-

market ratio, or depressed prices relative to historic levels—seem to have provided higher 

average returns than “glamour” or growth stocks. 

 One way to address the problem of data mining is to find a dataset that has not already 

been researched and see whether the relationship in question shows up in the new data. 

Such studies have revealed size, momentum, and book-to-market effects in security mar-

kets around the world. While these phenomena may be a manifestation of a systematic risk 

premium, the precise nature of that risk is not fully understood.